A large dataset of genotyped broilers (over 23,500 individuals, genotyped for 600k SNPs) with phenotypic records on body weight (BWT), feed intake (FI) and hen house production (HHP) was analysed to test the effect of training population (TRN) size and chip density on the accuracy of breeding value prediction. Using 4 cross-validation scenarios and 7 chip densities, the size of TRN was determined as a limiting factor. With large enough TRN, increasing chip density over 20k SNPs did not bring further increases to accuracy. On the other hand, increasing TRN size improved accuracy across all chip densities, most clearly so for BWT and FI.